Instrumentation and testing methodology. Conclusions on synthetic tests

The release of the new version of the DirectX API inspires hope for another revolution in graphics, steadily bringing it closer to the bright future promised seven years ago in the form of cinematic quality. Given the shift in priorities of igrostroev to console projects, we will not have to wait for the next PC masterpiece soon. In the meantime, GPU developers are on the defensive, releasing the second generation of desktop graphics cards with support for DirectX 11.

The transition to this API turned out to be difficult for NVIDIA: cards that were half a year late were expensive to manufacture, with high power consumption and some of the functional blocks of the company's flagship were disabled. Only now she was able to release more or less worthy solutions. AMD, on the contrary, is steadily following its strategy of developing simple single-chip products and high-end accelerators based on two GPUs. And if not for her, interest in multi-chip configurations would hardly have been at the proper level.

AMD decided to change the generations of video cards by expanding the Radeon line, thereby introducing confusion into the company's lineup. The novelties of the HD 6800 series actually turned out to be slower than their predecessors, but due to the price they should occupy a niche between the Radeon HD 5800 and HD 5700, competing with the good-looking GeForce GTX 460.

For high-performance solutions based on a single GPU, there is a place where two-chip products usually reigned. Thus, the Radeon HD 6900 series will now include all cards based on the Cayman core - the successor to Cypress.

Cayman

The Cayman graphics processor, on the basis of which the Radeon HD 6970 and HD 6950 video cards are based, is made according to the same 40-nm process technology as its predecessor. But unlike Cypress, the new kernel has undergone some changes. So, the number of SIMD cores was increased from 20 to 24 (16 superscalar stream processors in each), the number of texture units increased accordingly to 96. Render units were slightly improved to work with anti-aliasing, but their number remained unchanged, i.e. 32 pieces. The smoothing algorithm has also undergone some improvements. The "graphics engine" is now divided into two, each serving a different half of the core. It has doubled the number of some blocks, which made it possible to increase the speed of working with geometry and tessellation (the latter is almost three times faster than Cypress). In addition, each half of the kernel has its own thread manager (Ultra-Threaded Dispatch Processor), which is responsible for distributing the load among the executing units.

But the most important thing is that AMD abandoned the fifth ALU (T-unit) in superscalar processors, capable of executing a complex instruction (Special Function) - now its role is distributed among three of the remaining four. This approach, according to the developer, allows to increase productivity up to 10% per square millimeter of the crystal area.

In the end, the number of stream processors was 1536, while Cypress has 1600. But this is unlikely to negatively affect the performance of the new GPU, especially since its frequency has become slightly higher - 880 MHz versus 800 MHz. The GDDR5 memory frequency has also been increased - up to 5500 MHz, although the bus width is the same as 256 bits. But the volume of the video buffer, without hesitation, was made twice as large, and now it is equal to as much as 2048 megabytes, which will be especially useful at high screen resolutions and in CrossFireX mode.

The characteristics of the younger model are a little more modest: 1408 stream processors, 88 TMUs, the same number of ROPs, the core and memory frequencies are 800 and 5000 MHz, respectively. The amount of memory is also 2048 MB.

Video adapter	Radeon HD6970	Radeon HD6950	Radeon HD6870	Radeon HD6850	Radeon HD5870	Radeon HD5850
Core	Cayman XT	Cayman Pro	Barts XT	Barts Pro	RV870 (Cypress)	RV870 (Cypress)
Architecture	VLIW4	VLIW4	VLIW5	VLIW5	VLIW5	VLIW5
	2640	2640	1700	1700	2154	2154
Process technology, nm	40	40	40	40	40	40
Core area, sq. mm	389	389	255	255	334	334
	1536	1408	1120	960	1600	1440
Number of texture blocks	96	88	56	48	80	72
Number of render units	32	32	32	32	32	32
Core frequency, MHz	880	800	900	775	850	725
Memory bus, bit	256	256	256	256	256	256
Memory type	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5
Memory frequency, MHz	5500	5000	4200	4000	4800	4000
Memory size, MB	2048	2048	1024	1024	1024	1024
	11	11	11	11	11	11
Interface	PCI-E 2.1	PCI-E 2.1	PCI-E 2.1	PCI-E 2.1	PCI-E 2.1	PCI-E 2.1
Declared maximum power consumption, load/idle, W	(250) 190/20*	(200) 140/20*	151/19	127/19	188/27	170/27

* —in parentheses is the maximum power

Naturally, with an increase in the number of functional blocks with an unchanged technological process, the Cayman crystal area increased to 389 mm 2 - after all, it became “thicker” by 500 million transistors. In this case, we should expect an increase in the power consumption of new cards. But with the release of Cayman, AMD has changed the presentation of this indicator. Now, thanks to the new PowerTune technology, the frequency of the GPU, and not just its voltage, can dynamically change depending on the load and power consumption. The latter percentage can be configured independently in the Catalyst Control Center (ATI Overdrive tab).

At standard settings, the power consumption of the Radeon HD 6970 in normal gaming applications is 190 watts, while the peak in the OCCT benchmark can reach 250 watts. For the Radeon HD 6950, these values ​​are smaller and amount to 140 and 200 W, respectively. At idle, the cards consume only 20 watts.

If the slider in ATI Overdrive is moved to -20%, the GPU frequency will change within 880-500 MHz (with a higher load it will decrease, with a lower load, on the contrary, it will increase), thereby reducing the power consumption and heat dissipation of the card.

If you move it to + 20%, then instead of changing the frequency, a higher level of overclocking will be available to the user.

Of the pleasant moments for enthusiasts, we note Dual BIOS support and a special switch near the CrossFire connector that allows you to select one or another chip with microcode. One version is designed to be constantly updated, and the second is write-protected to prevent information corruption. This will be especially useful for those who like to reflash cards - if one microcircuit fails, then it will be possible to continue using the accelerator.

The cooling system of Radeon HD 69xx video cards is based on an evaporation chamber, like the Radeon HD 5970 and the newly minted GeForce GTX 580. Previously, this design was found in Sapphire products, and as practice has shown, the efficiency of such solutions is higher than that of standard heat pipe coolers.

The number of connectors for connecting monitors has increased and now, instead of one standard DisplayPort, two miniDP version 1.2 will be available to users. Eyefinity support has remained in place.

As you can see, the changes in the new Cayman GPU are enough to talk about a new generation of video cards with DirectX 11 support. But how will these changes affect real applications? This is what we will find out below.

Now let's take a look at the cards themselves. Before us are almost complete twins. Instead of the smooth lines and rounded shapes of the previous Radeon series, newcomers have a "brick" look with red accents on the sides.

Radeon HD 6970 is equipped with six- and eight-pin power connectors.

The younger Cayman has two six-pin sockets.

The reverse side of the video cards is covered with a black metal plate.

The rear panel has a pair of DVI and mini-DisplayPort connectors, plus one HDMI.

The cooling system of both models is the same.

The plate on the back of the board does not cool the memory chips, as was the case with the Radeon HD 5870 or Radeon HD 5970. All GDDR5 chips are soldered on the front side. The design of the turbine is quite typical and includes the main radiator for cooling the core and a massive metal base that plays the role of a heat sink for the memory and power elements of the power system. Only instead of traditional heatpipes, the reference cards based on the Cayman are equipped with a heatsink with a vapor chamber.

The native CO looks quite impressive.

The Foxconn PVB070G12N fan with a diameter of 70 mm is responsible for blowing the entire structure.

The younger card uses a different fan, but of the same diameter.

Let's take a look at the board design now. Below are photos of the Radeon HD 6970

And here is the Radeon HD 6950:

No difference! And with the Radeon HD 5870, the differences are minimal.

They differ in video cards with different GPUs and operating frequencies.

Radeon HD 6970 (left) and Radeon HD 6950

Both adapters are equipped with 2 GB of GDDR5 video memory. The older model uses Hynix H5GQ2H24MFR R0C chips, designed for an operating frequency of 6 GHz.

On the Radeon HD 6950, Hynix H5GQ2H24MFR T2C chips are soldered, the nominal value of which is 5 GHz.

The GPU-Z utility correctly determines the frequencies of new video cards, making mistakes only in the number of computing units.

Monitoring and overclocking in MSI Afterburner also works fine. So far, only there is no support for softvoltmod. The power converters on the cards are based on Volterra VT 1556MF controllers that support I2C and software voltage control is possible on them. So we expect this feature in future versions of Afterburner.

A little about temperature. The Radeon HD 6970 in gaming applications easily warms up to 90 ° C (at 23 ° C indoors). Almost like the GeForce GTX 480, only the noise level of the new AMD is much less, the hum is noticeable, but tolerable.

The Furmark stress test warmed the card up to 92°C, but the noise level was already very high.

Let's comment a little on the Power Control function, which allows you to switch the card to a more economical mode. With a value of -5%, we did not see any drop in performance, but with -20%, the core frequency jumped from 500 to 800 MHz and the result fell by 26% in the Crysis Warhead test.

Increasing Power Control helped a bit with overclocking. The card was able to work at higher frequencies, but it still failed to achieve stability at them. Plus, we also noticed a meager drop in performance in some benchmarks (up to 0.2%) when increasing the Power Control limit even at a slight overclock. As a result, we still stopped at a frequency of 965 MHz, which obeyed the core even at the nominal voltage. This is also a very good result. The memory was overclocked to 1565 (6260) MHz. Stability in this mode was possible only at increased turbine speeds.

The junior Radeon HD 6950 was cooler and quieter. In game mode, it did not warm itself above 80 ° C.

Even at Furmark, peak temperatures never exceeded 87°C.

From the nominal 800/5000 MHz, we managed to overclock this instance to 860/5640 MHz. The memory results can be explained by the use of other chips, but we expected more from the GPU.

Remembering the overclocking results of the older card and the same motherboard design for both Radeons, users can only wait for utilities that will allow them to control the voltage. Power Control in this case did not affect overclocking at all. Characteristics of the tested video cards

The AMD Radeon HD 5970, Force3D Radeon HD 5870, Zotac GeForce GTX 580, and Inno3D GeForce GTX 480 cards used in a recent GF110-based solution test were used for comparison with the AMD Radeon HD 6970 and HD 6950. For comparison, their characteristics are listed in the following table:

Video adapter	Radeon HD6970	Radeon HD6950	Radeon HD5970	Radeon HD5870	GeForce GTX 580	GeForce GTX 480
Core	Cayman XT	Cayman Pro	RV870x2 (Hemlock)	RV870 (Cypress)	GF110	GF100
Number of transistors, million pieces	2640	2640	2154x2	2154	3000	3200
Process technology, nm	40	40	40	40	40	40
Core area, sq. mm	389	389	334x2	334	512	526
Number of stream processors	1536	1408	1600x2	1600	512	480
Number of texture blocks	96	88	80x2	80	64	60
Number of render units	32	32	32x2	32	48	48
Core frequency, MHz	880	800	725	850	772	701
Shader domain frequency, MHz	880	800	725	850	1544	1401
Memory bus, bit	256	256	256	256	384	384
Memory type	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5	GDDR5
Memory frequency, MHz	5500	5000	4000	4000	4008	3696
Memory size, MB	2048	2048	1024	1024	1536	1536
Supported version of DirectX	11	11	11	11	11	11
Interface	PCI-E 2.1	PCI-E 2.1	PCI-E 2.1	PCI-E 2.1	PCI-E 2.0	PCI-E 2.0
Declared maximum power consumption, W	(250) 190	(200) 140	294	188	244	250

test bench

The test bench configuration is as follows:

• Processor: Core i7-965 (3, [email protected].95 GHz, BCLK 172 MHz);
• cooler: Thermalright Venomous X;
• motherboard: Gigabyte GA-X58A-UD3R (Intel X58 Express);
• memory: G.SKILL F3-12800CL8T-6GBRM (3x2GB, [email protected] MHz, 8-8-8-24-1T);
• hard drive: Hitachi HDS721010CLA332 (1 TB, SATA2, 7200 rpm);
• power supply: Seasonic SS-850HT (850 W);

• operating system: Windows 7 Ultimate x64;
• video card drivers: NVIDIA GeForce 260.99, ATI Catalyst 10.11.

User Account Control, Superfetch, Windows Defender, paging file and interface visual effects were disabled in the operating system.

Test Methodology

The games in our testing are divided into three groups, according to the version of DirectX supported. The first are applications for the ninth version of this API, and the last are games for DirectX 11. Within each group, applications are arranged in accordance with the release date, the exception is Metro 2033, this is the most demanding game, and we traditionally put the results in it at the very end of the article. All tests were carried out at 1920x1200 resolution at maximum quality settings, AF16x anisotropic filtering is always enabled. Anti-aliasing was activated from the game menu, forced forcing through the drivers was not used. The list of games and testing methodology are described in more detail in the GeForce GTX 580 review.

Test results in applications under DirectX 9

Borderlands

This game based on Unreal Engine 3 favors NVIDIA video adapters. The Radeon HD 6950 is slightly inferior to the Radeon HD 5870, but the HD 6970 even outperforms the dual-chip card in terms of minimum fps.

mafia 2

No more than 12% performance difference between Cayman cards. The older Radeon HD 6970 is 8% faster than the GeForce GTX 480.

The very low result of the Radeon HD 5970 immediately catches the eye. Multi-chip configurations are not supported by the game. The Radeon HD 6950 is a couple percent behind the HD 5870, but its older brother is on the heels of the GeForce GTX 580.

Test results in applications under DirectX 10

Nothing unexpected. The Radeon HD 6950 is a couple of percent behind the Radeon HD 5870, and the Radeon HD 6970 outperforms the GeForce GTX 480 by 9%.

Far Cry 2

The complete excellence of the Fermi architecture. You can also see a good advantage of Cayman over Cypress, when the overclocked Radeon HD 6970 almost overtakes the Radeon HD 5970.

Tom Clancy's H.A.W.X. 2

Familiar situation. Active use of tessellation confuses AMD cards, especially old solutions based on Cypress, even the Radeon HD 6950 is 27% faster than the Radeon HD 5870.

Sid Meier's Civilization 5

Unusual results for both GeForces - when overclocked, there was not an increase in the final result, but its tiny decrease to about 3000 points (these data are roughly rounded). In face value, there is little difference between them. But even with such a meager drop in results, NVIDIA cards left behind all AMD's single-chip rivals.

Based on the results of our testing, the new single-chip AMD flagships demonstrated superiority over the old Radeon HD 5870, but did not make any splash. Something more was expected from the new series. Polishing the architecture and fixing bugs allowed us to slightly increase performance, the rest was done by higher frequencies.

The performance of the Radeon HD 6950 is close to the Radeon HD 5870, in some places the newcomer loses a couple of percent, in some places it comes out ahead with the same tiny margin. The main advantage is noticeable in games that actively use tessellation (Lost Planet 2, H.A.W.X. 2). A good lead over its predecessor in Metro 2033 is partly due to the increased amount of memory - in other cases, the second gigabyte is not in demand.

The Radeon HD 6970 is already a more interesting product, it clearly outperforms the Radeon HD 5870. Although in some games under older versions of DirectX the difference between them is no more than 10%. But in DirectX 11 we are seeing a thirty percent increase. This is still not enough to compete strongly with the dual-chip Radeon HD 5970 and NVIDIA's flagship. The older Cayman successfully competes only with the GeForce GTX 480, although in DirectX 11 the first place almost always belongs to the old NVIDIA. However, no one took the lead - the new graphics cards are designed to create competition for the freshly baked GeForce GTX 570, and the company has the same Radeon HD 5970 as a rival for the GeForce GTX 580.

Rather, Cayman is nothing more than preparation for the transition to a thinner process technology, where new GPUs can acquire a large number of computing units with an acceptable level of power consumption. Now, of course, this is impossible to achieve. This is confirmed by almost the same number of stream processors and TDP level as Cypress. Rather, on the contrary, the Radeon HD 6970 resembles the GeForce GTX 480 with its hot temper, although, fortunately, it does not differ from other top-end video adapters in terms of noise characteristics. The main advantage of AMD's new products is the price. Their cost is quite attractive against the background of the same GeForce GTX 580. To compete in the top segment, the company will have to release a solution based on two Cayman. But that's another story...

Testing equipment was provided by the following companies:

• 1-Incom - G.SKILL F3-12800CL8T-6GBRM memory;
• AMD - video cards Radeon HD 6970, Radeon HD 6950 and Radeon HD 5970;
• Gigabyte - Gigabyte GA-X58A-UD3R motherboard;
• Inno3D - Inno3D GTX 480 video card;
• PCShop - Force3D HD5870-1GB (F5870P-G5-390-FS) video card;
• Syntex - Seasonic SS-850HT (S12D-850) power supply;
• — hard drive WD3200AAKS;
• Zotac - GeForce GTX 580 graphics card (ZT-50101-10P).

At the end of last year and almost the entire current year, AMD dominates the desktop graphics market with its Evergreen family of accelerators. Released in September 2009, products based on the Cypress GPU, as well as their junior modifications, have rightfully taken their place in the computers of many users. Naturally, the success of AMD accelerators cannot be called accidental. While NVIDIA was solving its production problems, trying to bring GF100-based accelerators to the market, AMD was already promoting its DirectX 11-compatible solutions with might and main in all segments of the desktop graphics market, actively winning market share from a competitor. Even the birth of the GeForce GTX 480 could not stop the “red locomotive”, since the new NVIDIA flagship was not as fast as the public wanted, in addition, it was noticeably hotter and more expensive than the Radeon HD 5870. Finally, quite recently, NVIDIA began to correct situation for the better. Serious work on the bugs, carried out by the company's engineers, allowed the new GTX 5xx accelerators to take their rightful place among the most productive desktop solutions for PCs. Meanwhile, AMD fans were rubbing their hands with malice, waiting for a powerful response from AMD, capable of returning the championship title among single-chip solutions to “red graphics”.

Today, the official announcement of the Radeon HD 6950 and Radeon HD 6970 video cards took place - it was they who, according to rumors, were perceived as a response to the release of the GeForce GTX 570 and GeForce GTX 580. Let's try to figure out how effective AMD's new weapon is, and how the company actually positions their new products.

So, contrary to the expectations of the public, AMD does not consider its new graphics accelerators Radeon HD 6970 and Radeon HD 6950 as direct competitors to the GeForce GTX 580 and GeForce GTX 570. As opposed to the fastest NVIDIA accelerator, AMD still puts the Radeon HD 5970, which, as our testing does not always cope with the mission assigned to it. In turn, the Radeon HD 6970, according to AMD, should be faster than the GeForce GTX 570 and slightly slower than the GTX 580, but the Radeon HD 6950 has a special role - it is believed that the new product should create its own niche in the hierarchy of modern 3D accelerators.

When developing new GPUs, AMD engineers faced a number of tasks:

• Creation of an efficient architecture for graphical and non-graphical computing;
• Achieving high geometric performance;
• Implementation of new picture quality modes;
• Implement effective nutrition management practices.

Since the advent of the R600 graphics processor, on the basis of which the HD 2900 series accelerators were built, then ATI, and now AMD, has been using the so-called VLIW architecture (Very Long Instruction Word - Very Long Machine Word).

Cypress GPU Block Diagram

Block diagram of GPU Barts

GPU Cayman Block Diagram

The essence of this architecture is the parallel execution by the kernel of several operations at the same time. Until now, ATI/AMD GPUs have used the so-called VLIW5 architecture. This meant the following: the Cypress GPU contained 20 SIMD cores, each such core contained 16 blocks of superscalar stream processors (Stream Core).

The internal structure of each stream processor is as follows:

• Four execution units + Special Function Unit (SFU or Special Function Unit);
• Branch block;
• Block of general purpose registers.

It turns out that there are 20 * 16 = 320 stream processors inside the Cypress chip, and given the structure of each stream processor, we can assume that the Cypress GPU contains 1600 (320 * 5) scalar 32-bit stream processors. Thanks to this device, the Cypress GPU has a peak processing power of up to 2.7 Tflops in single precision calculations and up to 544 Gflops in double precision (FP64) operations. The used VLIW-architecture eliminates a number of complex nodes inside the GPU, which makes AMD graphics chips quite economical, especially in comparison with top competitor products. However, not everything is as smooth as it seems at first glance. Parallel execution of computational operations often causes blocks to stand idle waiting for the execution of one or another operation, when the result of one calculation depends on the result of another. Manual programming for chips with VLIW architecture requires a lot of effort, since the programmer needs to take into account a huge number of internal dependencies of code threads, as a result of which the efficiency of computing on such GPUs largely depends on the optimization of compilers. Cayman GPUs, which are the heart of the Radeon HD 6950/6970, received a redesigned VLIW architecture, the so-called VLIW4.

Using VLIW4 allows you to better optimize the load of stream processors, improve the functionality of each execution unit (Stream Processing Unit - SPU) and get rid of the special functionality unit (SFU (Special Function Unit)), which was responsible for calculating the so-called transcendental functions. As a result, according to AMD, Cayman graphics chips have increased efficiency per 1 sq. mm. crystal area. Thanks to the transition to VLIW4, it was possible to improve the core design, increase the number of SIMD cores, texture units, and also rework a number of other GPU units.

Improved ROP efficiency with improved write pooling (more efficient memory bus loading)

The Cayman GPU is the first to implement “Asynchronous dispatch” technology. Previously, the graphics core had one common command queue, so computing and graphics tasks lined up in this queue according to the “first come, first served” principle. This is like a single core CPU with many ALUs but only one decoder. For the first time, the Radeon HD 6900 series supports asynchronous processing of many independent command queues. Each queue has its own priority and its own virtual memory. Two independent full-speed DMA (Direct Memory Access) controllers are also available for simultaneous reading and writing of system memory via the PCI-E bus. In addition, double-precision calculations have reached ¼ of the peak speed of single-precision calculations.

It's no secret that AMD has been criticized for a long time for its low geometry processing speed and slow work with tessellation. According to the developer, the geometry processing speed has doubled, and when tessellation is enabled, in some cases, an almost threefold increase in performance is achieved compared to the Radeon HD 5870. Such results became possible due to the increase in the corresponding functional blocks responsible for working with primitives, as well as the use of a more advanced tessellation block.

Finally, before moving on to an external examination of Radeon HD 6950/6970 accelerators, let's turn to a summary table with technical characteristics and prices.

Video card name	AMD Radeon HD 5870 1 GB	AMD Radeon HD 6850 1 GB	AMD Radeon HD 6870 1 GB	AMD Radeon HD 6950 2 GB	AMD Radeon HD 6970 2 GB
Kernel codename
Process technology, nm
Number of transistors, mln.
Maximum power consumption, W (PowerTune Maximum)
Typical power consumption, W
Power consumption at rest, W
GPU core frequency, MHz
Number of ROP blocks, pcs
Number of TMUs, pcs
Number of universal processors
Shader domain frequency, MHz
Video memory type
Memory bus width, bit
Effective video memory frequency, MHz
Video memory bandwidth, GB/s
Approximate retail price according to Market 3Dnews / recommended retail price c.u.

Based on the technical characteristics of the new products and mindful of all the architectural improvements in the Cayman, we believe that the AMD Radeon HD 6950 accelerator as a whole should be approximately on the same performance level as the Radeon HD 5870. Somewhere HD 6950 will be faster (in the case of tessellation - noticeably faster), somewhere slower due to the lower clock frequency of the graphics core. In turn, the Radeon HD 6970 should be ahead of the old Radeon HD 5870 in all applications, especially the latest DirectX 11 applications with tessellation support. If we talk about prices, there is no data on the retail price of the new accelerators yet, but it is already clear that the new AMD accelerators will be noticeably cheaper than the GeForce GTX 570/580, although, in our opinion, they are unlikely to reach the flagship level in terms of performance. NVIDIA accelerators. How much? Will show testing. In the meantime, let's move on to the external examination of the heroes of our review.

AMD Radeon HD 6950/6970 accelerators look practically the same from each other, even the size of the boards is identical. Accelerators can be distinguished only by the sticker on the casing of the CO, as well as by the power connectors. The Radeon HD 6950 accelerator is powered by two 6-pin PCI-Ex connectors, while the Radeon HD 6970 requires one 6-pin and one 8-pin connector. On the reverse side of the board is almost completely covered with metal plates. Apparently, this was done to evenly distribute heat over the surface of the accelerator.

The number and type of connectors on the output panel of the Radeon HD 6950 and HD 6970 are the same: two Mini DisplayPort version 1.2 connectors, HDMI 1.4A, 2x DVI (DL-DVI + SL-DVI). Compared to the Radeon HD 6850/6870, there are no differences in the port layout.

We dismantle the cooling system Radeon HD 6950/6970. The CO design of both video cards is completely the same. A massive metal plate is attached to the CO plastic casing, which is in contact with the memory chips and power system elements through special thermal pads. An aluminum heat sink is responsible for removing heat from the GPU, the sole of which is a copper evaporation chamber. Contact with the GPU occurs through a thin layer of thermal paste.

Radeon HD 6950 rear

Radeon HD 6950 Front

Radeon HD 6970 rear

Radeon HD 6970 Front

Only after the dismantling of the cooling system, it can be seen that the PCB of the accelerators are slightly different in color. The textolite of the Radeon HD 6970 accelerator has a brown tint, and the textolite of the Radeon HD 6950 is black. The power system of both cards is identical:

• Volterra VT1156MF - GPU voltage (6 phases).
• Memory - 1 phase.
• Vddci (memory controller) - 2 phases.
  

Both video cards are equipped with Hynix H5GQ2H24MFR memory. The memory capacity of each of the video cards is 2 GB, the nominal clock frequency is 6 GHz (QDR).

Introduction

This review is dedicated to the reference video card Radeon HD 6950 2048 MB manufactured by Sapphire, which belongs to the category "no longer high-end, but not yet mainstream". A little more than half a year has passed since the release of these video cards on the market, but they still have not lost their relevance. During this time, AMD Catalyst drivers have been optimized and many new games have appeared, some of which are very resource-intensive. For example, Crysis 2 with high-resolution textures in DirectX 11 mode can "drop" performance below the "playability" level on any single-GPU video card, even a well overclocked one. Today's testing of Sapphire Radeon HD 6950 MB will be carried out in this and eleven other games. It will also talk about different ways to turn on blocked stream processors and check the video card for overclocking, both air-cooled and liquid-cooled.

Specifications

Technical specifications of AMD Radeon HD 6950 and AMD Radeon HD 6970 are listed in the table:

Characteristic	Radeon HD 6950
GPU	Cayman Pro (RV970)	Cayman XT (RV970)
Process technology, nm	40	40
Universal Processors	1408
Memory size, MB	2048
Memory type	GDDR5	GDDR5
Memory bus width, bit	256
GPU frequency, MHz	800	880
Memory frequency, MHz	1250 (5000)	1375 (5500)
GPU voltage (2D/3D), V	0.90/1.10
Supply system		6 phase Vgpu + 2 phase Vmem + 1 phase Vddci
DirectX support	Direct X 11	Direct X 11
OpenGL support	Open GL	Open GL
Multi-GPU support	AMD CrossFireX (2-Way, 3-Way, 4-Way)	AMD CrossFireX (2-Way, 3-Way, 4-Way)
Maximum TDP, Watt	200
Power consumption in the load, Watt	150	190
Power consumption at rest, watts	20	20
Interface	PCI Express 2.0 x16	PCI Express 2.0 x16
Price, USD	$300	$370

Packaging, packaging, PCB design and features

The video card comes in a large black box, which briefly lists its characteristics and a list of supported technologies:

The package, in addition to the video card itself, includes the following set:

• Installation instructions;
• CD with drivers and software;
• One flexible bridge Crossfire;
• Two adapters for connecting additional power 1x Molex 4-Pin -> 1x PCI-E 6-pin;
• One mini-DisplayPort to DisplayPort adapter;
• One DSub adapter<->DVI;
• One HDMI cable<->HDMI;
• Coupon for registration in Sapphire Select Club and company sticker.

The video card is fully consistent with the reference AMD Radeon HD 6950. It uses the same cooling system, PCB design and nominal frequencies. The only difference is the sticker on the front.

The length of the video card, excluding the mounting bracket to the case, is 273 mm (or 286 mm, including the bracket). This is enough to completely cover the width of not only a standard (ATX) motherboard, but also an enlarged one (EATX). Therefore, it is desirable that the motherboard you are using does not have high heatsinks on the southbridge, and the SATA connectors are turned sideways.

The height of the video card is standard, almost does not exceed the height of the mounting bracket. There are no heat pipes protruding upwards or a printed circuit board increased in height, as is often the case on non-reference video cards. And in width, the video card occupies two slots on the motherboard.

On the bracket for attaching to the case there are holes for blowing out heated air and a set of external interfaces for connecting monitors (two DVI, two mini-DisplayPort and one HDMI):

On the top are two 6-pin PCI-E connectors for connecting additional power, two connectors for connecting two to four video cards in Crossfire mode, and a switch to select the active BIOS chip. By default, this switch is set to "1", which corresponds to the selection of the main BIOS chip, which is write-protected by software. If the user wants to experiment with the BIOS (for example, to enable disabled stream processors), then the switch will first need to be moved to position "2", which corresponds to an additional BIOS chip that is open for modification by the user.

This was done in order to exclude the possibility of bringing the video card to a non-working state by flashing an incorrectly modified BIOS image by the user. Now, in order to restore the “messed up” user BIOS, it is enough to switch to the main BIOS (1) until the next start of the computer, and after loading the operating system, switch back to the additional BIOS (2) and flash it again. This allows you to experiment with BIOS firmware from other video cards without consequences, as well as change parameters such as nominal frequencies, voltages, fan speeds and the upper overclocking limit in AMD Catalyst Control Conter.

The design of the graphics card is still designed to use two slots, like previous generations of reference AMD graphics cards. Even if you remove the bracket for attaching to the case and replace the cooling system with a single-slot one (for example, a thin water block), the dual non-removable block of two DVI connectors will still not allow you to use the adjacent slot on the motherboard.

The video card is equipped with the RV970 GPU, better known as AMD Cayman, manufactured in the 45th week of 2010 using the 40nm process technology. The division of the RV970 GPU into Cayman Pro (Radeon HD 6950) and Cayman XT (Radeon HD 6970) is conditional. Physically, this is the same chip with 1536 stream processors. The only difference is that in the Radeon HD 6950 they are partially blocked by software at the BIOS level (only 1408 out of 1536 are enabled).

The GPU is not covered by a lid on top, but an aluminum frame is glued on the edges. The height of this frame does not exceed the height of the GPU itself, so there is no need to remove it to install an alternative cooling system. Around the GPU in a square with a side distance of 53 mm (75 mm diagonally) there are four holes with a width of 3 mm for mounting the cooling system.

The GDDR5 video memory uses Hynix H5GQ2H24MFR-T2C chips in an FBGA package with a capacity of 2048 Mbps. All eight chips are located on the front side of the video card. The nominal operating modes for this memory are 900 MHz (3600 Gbps) with a voltage of 1.35V or 1250 MHz (5000 Gbps) with a voltage of 1.50V. On the AMD Radeon HD 6950 under load, the memory operates in the second of these two modes.

To store the main and additional BIOS, two PMC-Sierra Pm25LV010 chips with a capacity of 1 Mbit are installed on the video card. One on the front and one on the back.

The nominal frequencies of the Sapphire Radeon HD 6950 are 800 MHz for the GPU and 1250 (5000) MHz for the video memory:

Power system, cooling system

Supply system

When developing a reference design for Radeon HD6 series graphics cards 8 50/HD6 8 70 AMD is moving away from Volterra components for the first time in years in favor of CHiL Semiconductor, Anpec, Texas Instruments, and Infineon. One of the main advantages of this approach is a significantly lower heating of the power system and, as a result, the possibility of a quieter operation of the cooling system, as well as the absence of problems with cooling the power system when using alternative coolers and universal water blocks.

Initially, to build a power system for Radeon HD6 video cards 9 50/HD6 9 70 was planned to use the same components as on the Radeon HD6850/HD6870, but the resulting shortage of Texas Instruments CSD59901M DrMOS chips forced them to change plans again and return to using Volterra components, only newer models compared to those used on previous generations of AMD video cards.

This doesn't rule out future updated designs for these cards, but for now, all reference Radeon HD6950/HD6970 (as well as the recently released Radeon HD6990) use the same Volterra components with traditionally high heat levels.

Most of the elements of the AMD Radeon HD 6950 power system are concentrated on the right side of the video card on the front side:

The GPU power system (Vgpu voltage) consists of 6 phases and uses a Volterra VT1586 voltage controller, six Volterra VT1636SF DrMOS chips, and two Cooper Bussmann CLA1108-4-50TR-R and CLA1108-2-50TR-R choke assemblies.

And one more phase to power the memory controller (Vddci). The Volterra VT262WF controller used for this and the Cooper Bussmann FP1005R1-R15-R throttle are located on the left side of the video card.

You can monitor the temperature of the power system using the GPU-Z program (VReg Temperature on the Sensors tab).

Cooling system

The AMD Radeon HD 6950 cooling system is secured with fourteen screws. Ten of them hold the black-painted aluminum cover that covers the back of the graphics card. But on the reverse side there are no strongly heated elements, so there are no gaskets between the cover and the video card. This cover is not used for cooling, but for protection against chipped SMD elements and short circuits in case of installing several video cards close to each other or next to other boards.

From the inside, the cooling system contacts the GPU through the usual gray thermal paste and through thermal pads with all video memory chips, DrMOS chips and the Vddci voltage controller.

A copper evaporation chamber is used to cool the GPU. In its lower part, at the point of contact with the GPU, the surface, although not polished, is quite even.

From above, the radiator closes a plastic casing that restricts and directs the air flow.

And this air flow is created by a Firstdo FD9238U12D turbine with a diameter of 92 mm, powered by a voltage of 12V and consuming a current of 1.2A.

On the one hand, the evaporation chamber is soldered to a black heatsink that covers the video card over the entire surface, and on the other hand, thin aluminum fins are soldered onto it, through which air flows.

Like many of AMD's brick reference graphics cards, the Sapphire Radeon HD6950's cooling system can be very quiet, but only as long as the card is idle in 2D mode. When running 3D applications, the noise already becomes noticeable, but quite tolerable, and the temperatures rise to a level of about +90°C, which is already close to the thermal throttling limit. But the automatic speed control works correctly and does not allow the temperature to rise even higher.

The problem is that this level of temperature already leaves almost no margin for increasing voltage and overclocking. They can easily be knocked down by several tens of degrees by setting the fan speed to maximum and overclocking the video card even more, but using this mode constantly, to put it mildly, is uncomfortable.

The choice, as usual with references, is between "silence, high temperatures, low overclocking" and "noise, reduced heat levels and increased overclocking." There is no way to achieve silence, low temperatures and high frequencies from this system at the same time.

Software control of voltages and frequencies

You can control the frequencies of AMD video cards using the Overdrive function in the Catalyst Control Center, but in the case of Radeon HD 6950 video cards, it allows you to increase the frequencies to only 840/1325 MHz.

Plus 40 MHz to the GPU frequency is a mockery, not overclocking. And the memory on these video cards can easily be overclocked to frequencies of at least 1400, and if you're lucky, even higher than 1500 MHz. Of course, you can modify the BIOS using the Radeon BIOS Editor (RBE) to push the overclocking limit in the Catalyst Control Center higher, but it's better to set it to only + 20% to the AMD Power Tune limit and forget about regular overclocking tools from AMD. Fortunately, Overdrive is far from the only way to overclock AMD graphics cards.

Sapphire has developed a special TriXX utility for its Radeon HD 5xxx / 6xxx series video cards, which allows you to change the frequencies, voltage on the GPU, control the speed of the cooling fan and monitor the temperature of the GPU. It is available for free download on the Sapphire Select Club website, but will only work with Sapphire graphics cards. The definition of "own" video cards is implemented in the same way as in similar utilities from other manufacturers (ASUS Smart Doctor), by reading and comparing the Vendor ID in the BIOS. Therefore, the Sapphire TriXX can be easily fooled by flashing the Sapphire BIOS into any compatible video card from another manufacturer.

After starting Sapphire TriXX shows information about the installed video card on the Info tab:

This information, like the BIOS, can be saved to a file.

The top right corner displays the current GPU temperature.

On the Overclocking tab, you can control the frequencies of the GPU (from 300 to 1200 MHz) and video memory (from 400 to 1800 MHz), as well as change the voltage on the GPU for 3D mode (from 1.10V to 1.30V). There are four profiles for storing user settings.

This is quite enough for overclocking on air and liquid cooling. But if desired, the frequency limit can be increased by using several programs in turn. For example, first increase the frequencies in MSI Afterburner, and then in Sapphire TriXX.

The next Fan Control tab is for setting up the fan control of the cooling system. Three modes are available - automatic (control according to the algorithm set in the BIOS), fixed (hard-coded fan speed regardless of temperature) and custom (control according to the algorithm configured by the user). In the latter mode, you can set the limits within which the fan speed will change depending on the temperature.

In the last Settings tab, you can change the program settings:

• Run at system startup;
• Minimize the program window on startup;
• Restore frequencies at startup;
• Synchronize the settings of all video cards when using Crossfire;
• Show effective memory frequency (for example, 5000 instead of 1250 MHz);
• Install information "gadget" on the Windows desktop;
• Apply new frequencies immediately, without pressing the Apply button;
• Disable Ultra Low Power State (ULPS).

In addition to Sapphire TriXX, you can also use the universal MSI Afterburner to overclock the Radeon HD 6950:

The GPU frequency limit is the same as in Sapphire TriXX, and the video memory frequency is even slightly lower. The GPU voltage limit in the free version is 1.30V.

RivaTuner v2.25 can also work with Radeon HD 6950 after modifying the configuration file:

The AMD Radeon HD 6950, thanks to the use of the Volterra VT1586MF controller, supports much higher voltages (up to about 2 volts), but to install them, you need MSI Afterburner versions of Etreme or MOA Edition or modified by yourself. You can also work directly with this controller via the I2C interface. AMD Radeon HD 6950 "hangs" on bus number 6, and the device number of the Volterra VT1586MF controller is 70. To dump, you can use the command "MSIAfterburner.exe /i2cd6,70":

Four groups of registers are highlighted in red, responsible for the voltage on the GPU:

• Registers 94 and 95 (4B 00);
• Registers 96 and 97 (41 00) are the voltage for 2D mode (default is 0.90V);
• Registers 98 and 99 (55 00) are the voltage for 3D mode (default is 1.10V);
• Registers 9A and 9B (51 80).

In each of these groups, the first byte is responsible for setting the voltage with an accuracy of one hundredth of a volt (0.01V), and the second byte allows you to increase this accuracy to thousandths and even more. Thus, voltage 1.30V corresponds to VID-code "69 00", 1.305V - "69 80", etc.

To set the desired voltage, its VID code must be written to the corresponding registers of the Volterra VT1586MF controller. Support for I2C write commands has been added to MSI Afterburner since version 2.20 beta 5. But if they are executed on a Radeon HD 6950, nothing happens and the voltage remains the same. It can be assumed that this is either a "bug" in MSI Afterburner or these registers are specially write-protected.

BIOS Modification to Enable Disabled Stream Processors

As mentioned above, for the production of the Radeon HD 6950/6970, the same RV970 GPUs with 1536 stream processors are used. But on the younger video card, only 1408 are initially enabled, and the remaining 128 are blocked by software. The presence of a software, and not a hardware lock, suggests that this was done for a reason, but in order to increase the attractiveness of the younger model in the eyes of users. Not every buyer of the Radeon HD 6950 will be engaged in unlocking, but very many will know about this possibility. On the other hand, the manufacturer releases itself from liability, stating that unlocked video cards are not covered under warranty. At the same time, they protected themselves from inexperienced fans of "softvoltmod" by blocking the increase in voltage on the Radeon HD 6950, flashed using the BIOS from the Radeon HD 6970. Solid pluses, both for the manufacturer and for buyers.

There are two ways to enable locked stream processors on the Radeon HD 6950. Both of them require flashing the BIOS of the video card and, accordingly, will void the warranty on the video card. But even if it stops starting and is determined by software flashers, you can still use the programmer. Before proceeding with unlocking, do not forget to switch to the additional BIOS (2), available for user modification. Do not be afraid to experiment, in case of problems, you can always switch back to the main BIOS, restore the backup and continue working.

First way - BIOS flashing from AMD Radeon HD 6970. This method became known even before the video cards went on sale. It was first tested in practice by W1zzard from Techpowerup.com. And later, site visitors collected statistics on attempts to unlock video cards from different manufacturers, which showed that more than 90% of all Radeon HD 6950s were successfully unlocked. The Sapphire Radeon HD 6950 used for testing was no exception and retained the ability to work stable after turning on all stream processors.

In order to unlock your video card in this way, you first need to find and download a BIOS from a Radeon HD 6970 of the same manufacturer as your card and on the same PCB design. In the case of reference video cards, a BIOS from any other reference video card will do, regardless of the manufacturer. If you have doubts about the PCB design or a video card with 1024 MB of memory, you should use the other method described below.

The peculiarity of this method is that after flashing the BIOS from the Radeon HD 6970, along with the increase in stream processors, the nominal frequencies (from 800/1250 MHz to 880/1375 MHz) and the voltage on the GPU (from 1.10V to 1.15V) also change, which naturally leads to an increase in power consumption, heating of the video card and noise level. On the one hand, this is convenient if you still planned to slightly overclock your Radeon HD 6950, just up to the level of the Radeon HD 6970. On the other hand, using this method makes it impossible to program voltage control in programs such as Sapphire TriXX and MSI Afterburner.

After some time, a more perfect way to unlock was found and published - BIOS modification from Radeon HD 6950 using a special script Mod_BIOS_HD_6950 written in PHP by the same W1zzard from Techpowerup.com.

How to use it:

1. Download the archive with the Mod_BIOS_HD_6950 script and unpack it into a separate folder.
2. Save the current graphics card BIOS with ATIWinflash or ATIFlash and rename it to "original.bin"
3. Copy "original.bin" to the folder with the script.
4. Run "run.bat" in the folder with the script.
5. Flash the resulting "modded.bin" script into the video card using the same ATIWinflash or ATIFlash

You can check the success of unlocking using the GPU-Z program:

This method allows you to unlock only stream processors, leaving all other BIOS settings unchanged (frequency, voltage, etc.). After unlocking, there are no problems with the "softvoltmod". Works with any AMD Radeon HD 6950 graphics cards, including non-reference and 1024 MB memory.

If desired, instead of modifying and flashing the "native" BIOS, you can also try using the BIOS from other AMD Radeon HD 6950s. To do this, after downloading, rename it to "original.bin" in the same way and run the script.

Test configuration

For testing, an open stand was assembled with the following configuration:

• Processor: Intel Core i7-2600K D2;
• Motherboard: ASUS Maximus IV Extreme, Intel P67, BIOS 1902;
• Memory:
  • 3x2048Mb G.Skill Perfect Storm F3-16000CL7-6GBPS, DDR3-2000 (Elpida Hyper MNH-E);
  • 1x2048Mb Kingston HyperX KHX2000C8D3T1, DDR3-2000 (Elpida Hyper MGH-E);
• Video card: Sapphire Radeon HD 6950, 2048 Mb GDDR5, PCI-E;
• Drives: SSD Crucial m4 128 Gb (OS, benchmarks and games), HDD Western Digital WD1002FAEX;
• Power supply: Antec TruePower Quattro TPQ-1000, 1000W;
• CPU Cooler: Thermalright Archon with two Thermalright TY-140 fans;
• Thermal Grease: Arctic Cooling MX-4.

Software:

• OS: Windows 7 Enterprise SP1 x64 v6.1.7601 (english);
• DirectX Redistributable (Jun2010);
• Intel Chipset Device Software v9.2.3.1016;
• Intel Rapid Storage Technology Driver v10.6.0.1002;
• Intel HECI driver v7.0.0.1118;
• AMD Catalyst v11.8 Preview;
• CPU-Z v1.58;
• GPU-Z v0.54;
• Riva Tuner v2.25;
• MSI Afterburner v2.20 Beta 5;
• FurMark v1.9.1.

Air-cooled overclocking and temperature conditions

The room temperature during testing was +25°C.

MSI Afterburner v2.20 Beta 5 was used to overclock the video card. RivaTuner v2.25 was used to monitor GPU temperatures (of all blocks), and GPU-Z v0.54 was used to monitor the temperature of the power system. To create a load and warm up the video card, FurMark v1.9.1 was used.

To avoid power consumption throttling and frequency drops, the AMD Power Tune limit has been set to the maximum possible +20% in the AMD Catalyst Control Center. And to prevent this setting from being reset after applying frequencies in MSI Afterburner, the UnofficialOverclockingMode option in the MSIAfterburner.cfg configuration file was used.

To begin with, the operating temperatures of the video card were measured at standard frequencies. 800x1250 MHz.

In the mode with automatic fan speed control (30% / 1458 RPM at rest and 40% / 2125 RPM under load), the GPU temperature at rest was +62°C...+65°C, and under load - +90°C. ..+96°C. The power system under load warmed up to +63°C.

After setting the turbine mode to 100% (5700 RPM), the GPU temperature dropped by 20°C at rest (up to +42°C) and by 34°C under load (up to +54°C...+57°C). The temperature of the power system has dropped by 24°C (to +39°C).

The difference between the maximum and automatic modes of the cooler is huge, but not only in the obtained temperatures, but also in the noise level.

Now let's see how much the temperatures rise from overclocking the video card when compared in the quiet (automatic) mode of the turbine. Overclocking without raising the voltage, that is, at standard 1.10V, amounted to 900/1500 MHz. The GPU temperature rose quite a bit - by 4°C at rest (up to +65°C...+70°C) and by 3°C under load (up to +90°C...+98°C). The temperature of the power system has risen by 3°C (up to +66°C).

In acceleration at maximum turbine speed, it was possible to achieve frequencies 940/1520 MHz, but it is impossible to use it for permanent operation due to the high noise level.

We also checked the effect of turning on blocked stream processors on GPU temperature. Testing showed that there was no significant difference in temperature when using the BIOS modification method from Radeon 6950 (that is, without increasing the nominal frequencies and voltage).

Water-cooled overclocking and benchmark results

For further overclocking, a Topmods copper water block was installed on the GPU, and aluminum Zalman radiators were installed on the memory chips. And on the elements of the power system through a thermal pad and with the help of a flexible insulated wire, an aluminum radiator from the ASUS Radeon HD 6850 DirectCU video card was installed. To cool the water block, running cold water with a temperature of about +10°C was used.

The video card remained stable up to frequencies of 1025/1525 MHz with a voltage of 1.30V. Most of the benchmarks could be passed at 1050/1550 MHz with 1.33V, and 3DMark Vantage was once passed even at 1085/1585 MHz.

The results are as follows:

• 3DMark11 - Extreme Preset: (#1)
• 3DMark11 - Performance Preset: (#1)
• 3DMark11 - Entry Preset: (#1)
• 3DMark Vantage - Performance Preset: (#1)
• 3DMark2001SE: (#3)
• Aquamark3: (#2)
• Unigine Heaven - DirectX11: 1941.63 (#2)

Gaming performance

To minimize the dependence of the results on the speed of the processor, it was overclocked to a frequency of 5000 MHz with a voltage of 1.51V. Overclocking was carried out by increasing the multiplier to x50.

The AMD Catalyst driver settings have been left at default except for the following:

• Catalyst A.I. = high quality
• VSync=Always Off
• Power Control Settings = +20%

The following games were used to measure performance:

• Crysis 2 v1.9 - DirectX 11, Adrenaline Crysis 2 Benchmark Tool (Central Park);
• Metro 2033 v1.2 (Update 2) - DirectX 11, built-in benchmark (Frontline);
• S.T.A.L.K.E.R.: Call of Pripyat v1.6.02 - DirectX 11, built-in benchmark (SunShafts);
• DiRT 3 v1.1 - DirectX 11, built-in benchmark (L.A. Coliseum);
• Lost Planet 2 v1.1 (v1.0.1.129) - DirectX 11, built-in benchmark (Test B);
• Just Cause 2 v1.0.0.2 (Update 1) - DirectX 9, built-in benchmark (Concrete Jungle);
• Grand Thief Auto IV: Episodes From Liberty City v1.1.2.0 - DirectX 9, built-in benchmark (The Lost and Damned);
• Far Cry 2 v1.03 - DirectX 10, Far Cry 2 Benchmark Tool (Ranch Small);
• Total War: Shogun 2 v2.0 (v1.1.0 build 3409.285940) - DirectX 11, built-in benchmark (Sekigahara);
• Aliens vs. Predator DX11 Benchmark v1.03 - DirectX 11;
• Mafia 2 v1.0.0.4 (Update 4) - DirectX 9, built-in benchmark;
• Sid Meier's Civilization V v1.0.1.348 - DirectX 11, built-in benchmark (Askia).

Settings in all games were set to maximum quality. The only exception is Mafia 2. We had to disable the Apex PhysX option in it to avoid a large spread in results from run to run and limit the performance of the processor used to calculate physical effects when using this option with AMD video cards.

The tests were only carried out with a screen resolution of 1920x1080. Where possible, full-screen anti-aliasing was set to MSAA 4x mode, and anisotropic filtering to x16 mode. Vertical sync has been disabled. All games used either the built-in benchmark or third-party utilities to call the game's built-in FPS tools.

Performance was measured in four modes:

• Rated frequencies without stream processors enabled: 800/1250 MHz, 1408 SP
• Rated frequencies with the inclusion of stream processors: 800/1250 MHz, 1536 SP
• Acceleration on a standard cooling system with automatic turbine speed control: 900/1500 MHz, 1536 SP
• Overclocking on running cold water: 1025/1525 MHz, 1536 SP

Measurement results are shown as minimum and average FPS. Or only average, when the built-in game tools do not allow you to get information about the minimum FPS (Just Cause 2, Mafia 2). For games that allow you to get statistics on the construction time of all frames (frametimes), there are also graphs showing the distribution of FPS over the time it takes to pass the benchmark.

Testing has shown that there is a benefit to enabling disabled stream processors on the Radeon HD 6950, but it is not great. Raising the frequencies to at least the nominal level of the Radeon HD 6970 will give approximately the same increase in performance, and overclocking will allow you to squeeze even more out of the video card than by turning on stream processors. Therefore, this feature of the Radeon HD 6950 should be considered only as a small free bonus. Where the speed of the video card is not enough (Crysis 2 in DirectX 11 and Metro 2033 mode) to play with maximum settings, it will not be enough even after unlocking with overclocking.

Conclusion

Advantages and disadvantages of reference video cards AMD Radeon HD 6950 in general and Sapphire Radeon HD 6950 in particular:

[+] The ability to software unlock stream processors, which allows you to get an analogue of the Radeon HD 6970 for a lower price. There are also no problems with overclocking to the nominal frequencies of the Radeon HD 6970 (and even higher).

[+] Good performance/price ratio. Even now, half a year after entering the market, the Radeon HD 6950 2048 MB has no direct competitors in its price category. At the bottom there is an offer of the GeForce 560 Ti 2048 MB, which is slightly cheaper and slower than even the unmodified Radeon HD 6950, and on top - the GeForce 570 1280 MB, which has less video memory, higher power consumption and approximately equal to the Radeon HD [email protected] performance.

[+] Support for AMD Eyefinity technology, which allows you to use image output on multiple monitors simultaneously.

[+] Two gigabytes of memory for texture storage and a screen buffer allow you to use any screen resolution and the highest quality textures in all modern games. This is more than the reference GeForce GTX 570 and GeForce GTX 580.

[+] The ability to programmatically control the voltage on the GPU. To do this, you can use not only the universal MSI Afterburner, but also the proprietary Sapphire TriXX utility.

[+] Relatively quiet cooling system (if the video card is not overclocked, especially with raising the voltage), blowing heated air out of the case.

[+] Good equipment, the presence of an HDMI cable and a mini-DisplayPort adapter<->displayport.

[-] The presence of an artificial power consumption limiter in the form of AMD Power Tune technology. Forced frequency reset when power consumption exceeds a hard-coded limit. This greatly hinders overclocking, especially extreme. You can push the limit 20% higher with the corresponding option of the Catalyst driver, but for overclocking above 1050...1100 MHz and voltage above 1.30V...1.35V this may still not be enough. In addition, if you overclock the video card not using the driver itself, then it resets the set Power Tune limit at each frequency setting.

[-] The reference cooling system leaves almost no temperature margin for overclocking. You have to make a choice between good overclocking with low temperatures and a quiet operation of the video card.

[-] Poor tolerance to sub-zero temperatures. Artifacts and instability appear, even at temperatures achievable using a weak freon. Because of this, the vast majority of Radeon HD 6950/6970 video cards cannot overcome the overclocking level of about 1100 MHz.

[-] The power system is built using Volterra components, which require active cooling for normal operation. The standard cooling system copes with this task quite well, but when using alternative coolers or universal (non-fullcover) water blocks, you will have to at least install heatsinks on all DrMOS chips (Volterra VT1636SF) and it is desirable to provide at least a small airflow to these heatsinks.

And the Radeon HD 6970 presented at the beginning of this year are indeed the most effective and popular solutions of today. Their popularity is due to the competent pricing policy on the part of AMD and the presence of many progressive technologies such as AMD EyeFinity, which are only at the "rudimentary" level in the competitor's graphics solutions.
A large number of reviews of video cards of this family have already been presented on our Mega Review portal. Among them are:
1. PowerColor Radeon HD 6950 PCS++ review. Dual clock graphics card
2. Review of Sapphire Radeon HD 6950 2GB. Getting to know the reference video card
3. Review of HIS Radeon HD 6950. Unlock and turn into HIS Radeon HD 6970
4. Review of XFX Radeon 6950 1 GB XXX Edition. How will reducing video memory affect performance?
5. Review MSI Radeon HD 6950 Twin Frozr II. Modified version of the popular solution
6. Review of PowerColor PCS+ Radeon HD 6950 Vortex II Edition. New cooling system for the popular graphics card
7. Review of ASUS Radeon HD 6950. Modest factory overclocking from ASUS
8. Review of ASUS Radeon HD 6970 DirectCU II. Innovations from ASUS in the top graphics card from AMD
9. Review of HIS Radeon HD 6950. Unlock and turn into HIS Radeon HD 6970
10. PowerColor Radeon HD 6970 PCS+ review. Factory overclock and new cooling system
11. Review of PowerColor Radeon HD 6970 LCS. Trying to fool an overclocker?
12. Review of HIS Radeon HD 6970 IceQ Turbo. The new modification is already on the market!

Each of these reviews can be found on our Mega Review portal. When considering them, you can see that the difference between all graphic solutions lies solely in the modified cooling system and the level of factory overclocking of the core or video memory chips. All this makes part of testing video cards of the same type and of little interest to the end user.

Given this circumstance, we would like to prepare a different review of these products, which will summarize our experience with these video cards and allow end users to solve many problems that might have seemed unsolvable to them before.
We would like to build today's article in the framework of the "Question and Answer" format, which will make it easier to find the right information in the future. 1. What is the difference between AMD Radeon HD 6950 and HD 6970 graphics cards?
This is the first question that users ask when choosing a video card of this family. The difference between video cards is as follows:
- the younger video card has a smaller number of universal processors (instead of 1536 for the older one, the younger one has 1408 pieces),
- the younger video card has fewer texture units (instead of 96 for the older one, 88 for the younger one)
- the younger video card works at lower operating frequencies (the frequencies for the older video card are 880/1375 MHz, for the younger 800/1250 MHz),
- the younger video card has a lower power consumption (the older video card consumes up to 250 watts at peak load, the younger one - up to 200 watts of electricity).

The test results show that the objectively real difference in performance between the older and younger video cards can reach 17%. On average, this parameter fluctuates around 12%. It is up to you to decide if it is worth overpaying for an older video card of the family or not.

Image is clickable --

2. What is the real power consumption of video cards and will my power supply be enough for them?
This question is asked by our users in the second turn. In principle, the design of printed circuit boards for Radeon HD 6970/6950 video cards does not have any fundamental differences, but these differences are present in the power circuits of video cards. At least note the fact that the older video card is equipped with one six-pin and one eight-pin PCI-Express connector, and the younger one only with two six-pin PCI-Express connectors.

Despite this, the length of printed circuit boards is the same and reaches 27.6 cm, which must be taken into account when choosing a system unit case. Not every system unit of the mATX format will be able to accommodate a graphic solution of this family.

The manufacturer recommends a power supply unit with a capacity of about 550 watts for a single video card of this family. In this case, it should be borne in mind that this should be the real power of the power supply, and not the peak. The power supply must comply with the ATX2.xx specifications, that is, it must provide the main power through the 12 volt line.

If you plan to overclock a video card with a softvolt mod, then we recommend that you think about a higher power supply, since the power consumption of these video cards can easily approach 330-340 watts.

For a tandem of video cards of the Radeon HD 6950/6970 family in CrossFire mode, it is recommended to purchase a power supply unit with a capacity of at least 850 watts. As a rule, this power is enough to supply power to these graphic solutions.

Image is clickable --

3. What are the motherboard requirements for these video cards?
Experience shows that video cards of the Radeon HD 6950/6970 family work easily and without problems in all existing motherboards with PCI-Express slots. Officially, video cards support the PCI-Express 2.0 bus, which is backwards compatible with buses of previous generations, up to 1.0.

In general, using a graphics card in a PCI-Express 1.0 slot may result in the inability to achieve maximum performance in some game modes that require the fastest data transfer. Using a tandem of video cards in CrossFire mode on the bus of the previous generation is also not recommended, since its performance for two video cards is definitely not enough.

To create CrossFire configurations, you need to purchase motherboards that report support for this technology or have two or more PCI-Express 2.0 slots. Expensive motherboards support two full-fledged PCI-Express 2.0 x16 slots, cheaper options provide two video cards in PCI-Express 2.0 x8 format.

Experience shows that the difference in performance level between two different CrossFire configurations is insignificant, although it can be observed at high loads.

To build CrossFire, it is recommended to connect video cards with appropriate bridges. Ideally, these are two bridges. Experience shows that one bridge can be used if you play at a resolution of less than 1980x1024 pixels inclusive.
4. What brand of video card do you recommend to buy?
This is the next practical question that is beginning to interest our users. Indeed, today the domestic market offers a wide variety of Radeon HD 6950/6970 series products from ASUS, MSI, Palit, Gainward, Sapphire, PowerColor, HIS, etc. Most of the presented solutions have a reference design, which can be identified by the standard cooling system and power circuit wiring on the back of the video card.

If in your case you choose between reference designs, we recommend purchasing the cheapest option. The level of performance and quality of these video cards are absolutely the same. The difference between the solutions lies only in the labels applied to the casings of the cooling system.

Image is clickable --

If you want to purchase an alternative design graphics card, then the final choice is yours. More than 12 reviews of video cards of alternative design are presented on our Mega Review portal. The main differences between graphics cards other than the reference design are:
- at operating frequencies,
- cooling system,
- graphics card power supply.

As a rule, manufacturers install more efficient cooling systems on their video cards and carry out their slight factory overclocking. A striking example is the MSI Radeon HD 6950 Twin Frozr II video card, which has a modified cooling system and operates at higher frequencies than the nominal one.

Recently, a new modification of video cards of this series has entered the market - MSI Radeon HD 6950 Twin Frozr III PE, which has a modified power scheme. As a rule, modification of the power circuit of a video card leads to an increase in the stability of its operation during its overclocking and allows you to achieve higher frequencies than on a conventional video card. Unfortunately, no one guarantees this either, and this is only a theoretical "maybe", for which the user is invited to pay a little extra.

Image is clickable --

5. What technical problems have been identified with video cards at the moment?
Today, the technical problems of AMD Radeon HD 6950/6970 video cards can be associated with the appearance of an annoying "whistle" at high loads on the graphics solution. As a rule, these are the elements of the video card power system - chokes. The main reason is the lack of power of the power supply or the drawdown of the supplied voltage of 12 volts. Drawdown leads to an increase in current demand, which increases the strength of the current flowing through the chokes, which leads to the appearance of annoying squealing.

Unfortunately, the problem is not always in the power supply, it can also be in the video card itself. No one is immune from desoldering a somewhat poor-quality choke. This problem is solved quite simply. By ear, a whistling throttle is detected, a video card is removed and the identified element is filled with varnish for radio electronics.

It should be understood that this manipulation completely and forever deprives you of the warranty on the video card. As a rule, a whistling throttle is not recognized as a warranty case by any of the service centers.

There may be a simpler solution to this problem. Experience shows that the level of performance of video cards Radeon HD 6950/6970 is enough for many, and most of the computing power of video cards is wasted. Therefore, it makes sense to limit the maximum frame rate by enabling the VSync function in the driver. Reducing the load on the video card will lead to a decrease in both power consumption and the disappearance of annoying sound.

In general, Radeon HD 6950/6970 video cards are guaranteed no more often than competing solutions and even less, since they have two BIOS chips on board and users need to try to destroy both boot system records.

We also did not meet any complaints about the incompatibility of video cards with equipment from other manufacturers.

Image is clickable --

6. Why is the overclocking potential of Radeon HD 6970 video cards higher than that of Radeon HD 6950?
Indeed, both video cards are based on the same core. It's just that the younger model has a stripped-down core combined with lower power consumption, which is ensured by the difference in voltages applied to the core in 2D and 3D modes. Also, the video memory chips of the Radeon HD 6970 video cards work at a voltage of 1.6 volts, and for the younger Radeon HD 6950 video card at a voltage of 1.5 volts.

In principle, no one bothers you to increase this voltage. This can be done by uploading a new BIOS version to the video card. Some users have noted chip burn-in after increasing the voltage on the video memory chips. As experience shows, the number of these cases does not exceed the average values ​​even among non-modified video cards, so 1.6 volts for memory chips can hardly be called fatal.

Image is clickable --

7. What drivers do you recommend using with these video cards?
We do not have any preference and try to always use the latest drivers. As a rule, the latest versions of drivers can be found either on our Mega Review website or on the official AMD website.

The only key requirement is to first clean the operating system from the "remnants" of the drivers of the previous video card, especially if it was an NVIDIA video card. This is done quite simply, several programs are presented on the network and on our Mega Review website, for example, Driver Sweeper. Conclusion
Concluding our first article on what you should pay attention to when buying AMD Radeon HD 6950/6970 video cards, we want to note that these video cards are very popular among domestic users. Almost the only drawback of video cards of this family is the lack of support for Nvidia PhysX technology, which is why some users give unconditional preference to competitor products. Enthusiasts solved this problem a long time ago by installing a second graphics card from NVIDIA to support this technology in games. At the same time, NVIDIA PhysX will work without problems only in Windows XP and Windows 7 operating systems, the problem of activating this function in Windows Vista could not be solved.

Screenshots of the use of new AMD drivers are circulating on foreign sites, which include the work of PhysX on AMD Radeon HD 6950/6970 video cards. Unfortunately, no one has seen these drivers in our country, and all this still remains simple screenshots.

In the second article dedicated to AMD Radeon HD 6950/6970 video cards, we will try to give advice on the proper operation of these video cards.

The Radeon HD 6970 is an AMD graphics card released in December 2010. Its architecture simplified the design of stream processors and made them more efficient.

Model History

In 2008, AMD developed and released the Radeon HD 4000 series graphics card, codenamed R700 and branded by ATI Technologies. Probably many still use it in their PCs. The R700 architecture was replaced by the long-awaited R800, which paved the way for the 5000 lineup. The crown jewel of the series was the card that appeared in September 2009. It consisted of 2150 million transistors and was a success, remaining the flagship single-processor model for about 15 months. Replaced at the end of 2010. At the time it was one of the best graphics cards in its price range, which speaks for itself.

It took Nvidia about 6 months to fire back with the 470 and 480, and even then, many felt it fell short. After several quick price cuts and improved driver support, the card was made competitive. The continued introduction of the GTX 460 and the refined architecture of the GF110 helped create a viable product and paved the way for the next models, namely the GTX 580 and 570.

Meanwhile, AMD has been offering the HD 6000 since October 209, passing the baton to the HD 6870 as a counterbalance to the then $240 GTX 460. It was an impressive model, apart from a confusing naming policy and constant reminders from the manufacturer that it was not a replacement for the 5870. Starting with this series, the ATI Technologies brand was officially done away with for the purpose of correlation between AMD graphics products and its computer platforms. Accordingly, the design of the logo has also changed.

After a short delay, the new Radeon HD 6970 2GB was released. The graphics processor, codenamed Cayman, is made up of 2640 million transistors, which is 23% more than the 5870th model. Like the GTX 570 and 580, which are improved versions of the 480, the HD 6970 is an upgraded version of the 5870.

6900 Series Updates

The high-end Cayman architecture used in the HD 6900 series is slightly different from the Barts, which debuted in the 6870 and 6850 models. It used the VLIW5 HD 5000 configuration, which includes a SIMD with 4 simple and 1 complex flow processing units. Video cards of the 6900 series are distinguished by the VLIW4 configuration, in which stream processing is organized into groups of 4 blocks with common registers. Although they all have equal capabilities, 2 of them (3rd and 4th) perform special functions. According to information from AMD, the VLIW4 configuration provides processing power equal to VLIW5 while reducing die area by 10%.

Cayman processors offer a high level of parallelism compared to the Evergreen/Cypress architecture used in the HD 5800. Barts GPUs sit somewhere in the middle as they were a step up from Cypress in assigning individual thread managers to each of the two SIMD blocks. Cayman chips are even more powerful thanks to two GPE engines and the assignment of each to a SIMD block. This means that the architecture now has two more efficient tessellation units compared to Barts' one. For HD 6900 graphics cards, therefore, tessellation performance can be greatly improved. According to AMD, it's up to 3x the HD 5870. In addition, the Cayman architecture also features a redesigned rendering engine consisting of 128 Z/Stencil ROPs, 32 color ROPs, nearly twice as fast 16-bit integer operations, and 2-4 -fold increase in the speed of 32-bit floating point operations.

Cayman architecture

The transition from VLIW5 to VLIW4 has improved AMD's computing capabilities: narrower SPUs are easier to use to their full potential, FP64 improves performance by a quarter over FP32, and space savings can accommodate additional SIMDs. But if Cayman was a serious attempt to establish itself in the graphics computing market and grab a piece of the NVIDIA pie, then it had to do more than just add new shaders. Therefore, AMD has struggled to improve the functionality of its GPU so that it can pose a threat to the Fermi architecture.

The main feature of Cayman is asynchronous dispatching. The term very accurately describes what the card does. With Fermi, NVIDIA introduced support for parallel cores, which made it possible to compute multiple pieces of code at the same time. AMD repeated NVIDIA's approach, but went even further.

A design limitation of Fermi is that although the architecture allows multiple cores to be used simultaneously, each core must be served by a single stream processor. Independent applications, for example, cannot issue their own code and execute it in parallel, and the GPU must context-switch between them. AMD's Asynchronous Dispatch is designed to allow independent threads and applications to generate code that runs in parallel. At least on paper, this provides a significant advantage (context switching is too expensive), which should have surpassed the performance of Fermi.

The principle of asynchronous dispatching is that the GPU hides some of the information about its real state from applications and code, which, in essence, leads to the virtualization of GPU resources. After all, each piece of code believes that it is running on its own GPU with its own command queue and its own virtual address space. This puts the burden on the GPU and drivers, but the payoff is that it's better than context switching.

Asynchronous dispatch requires API support. DirectCompute is a fixed standard and does not support this feature, at least in version 11. Therefore, asynchronous dispatching is implemented as an OpenCl extension.

The rest of AMD's improvements are in memory and cache performance. The basic architecture here remains the same, but some minor changes have been made to the calculation. The local data store provided with each SIMD can now bypass the cache and global data store hierarchy by direct reading. Cayman received a 2nd direct memory access module, which improved the speed of reading and writing, allowing 2 operations to be performed simultaneously in each direction.

Finally, shader reading has been slightly sped up. Compared to Cypress, Cayman can reduce the number of operations by concatenating them.

Design

The length of the card is 27 cm, which is typical for modern high-performance models. For example, the HD 4870 X2 is 28 cm long, the same as the HD 5870. The dimensions of the HD 6970 are identical to the GTX 580. Like its predecessor, the AMD Radeon HD 6970 is manufactured using a 40 nm process, but the company has added 486 million more transistors, as a result whereby the crystal size increased by 16%. The GPU core runs at 880MHz, which is 3.5% faster than the HD 5870, and GDDR5 is also slightly faster at 1375MHz. The memory clock coupled to the 256-bit bus gives the HD 6970 a theoretical bandwidth of 176 GB/s, a 14.5 percent advantage over the HD 5870.

The HD 6970 also differs from the older 5870 model in its basic configuration. If the latter had 1600 cores, 80 TAUs and 32 ROPs, then the former received 1536 stream processors, 96 texture units and the same 32 ROPs (4% fewer cores and 20% more TAUs).

According to user feedback, peak integer bilinear texture filtering runs at 84.5 gigatexels/s and pixel fill at 28.2 Gp/s. Performance is 2.7 Tflops. A three-dimensional scene is built at a speed of 880 million polygons / s. The card supports dual streaming of HD video and has a built-in audio controller.

Inside the case is a printed circuit board with a 389 mm 2 GPU in the very center. Around the processor are 8 2GB Hynix GDDR5 memory chips with a nominal data transfer rate of 6Gb/s, which is 0.5Gb/s higher than the operating frequency of the card. The use of GDDR5 at high speeds is complicated by the difficulties in creating a good memory bus, which also affected this model. AMD has made progress towards 5.5 Gb/s through improved PCB design, but further speed increases seem to be counter-intuitive, if only because of the design of the 256-bit bus.

Cooler

The processor cooling is a fairly large aluminum heatsink of the evaporation chamber, composed of 39 plates 13.5 cm long, 6.5 cm wide and 2.5 cm high. The evaporator design was first implemented in the AMD ATI Radeon HD5970 and borrowed by NVIDIA in the GeForce GTX 570 and 580. Finally, cooling the heatsink is a 75 x 20mm fan that draws air in from the case and pushes it out at the rear of the card.

According to user reviews, for the most part, the cooler is very quiet, helped by low consumption (20 W) in idle mode. During the game, the fan, of course, accelerates, and the card consumes up to 250 watts under load. This is 33% more than the Radeon HD 5870 requires, but even with an increase in thermal load, the noise level does not rise to unacceptable levels.

The heatsink and fan are housed in a specially designed case that hides the entire graphics card. This is AMD's usual practice when designing its most high-end models. According to user reviews, they really like this design, as it provides reliable protection for the device. NVIDIA has also done this in the past for its most expensive graphics cards, such as the dual-GPU GTX 295, although its other flagship products (such as the GeForce GTX 580) have dropped this development.

Connectivity

In order to provide enough power to the card, AMD has installed 8-pin and 6-pin PCIe connectors. A similar solution can be found on the HD 5970 and GTX 580, since this configuration is usually used in models that consume a lot of power. Naturally, the HD 6970 supports CrossFire, and therefore a pair of connectors for connecting two or more cards can be found in the standard place. Next to them is a switch that allows you to choose between 2 BIOSes. This is done to improve the reliability of the model in case of failure of the flash memory. It is also possible to run the GPU using the backup system and then switch back to the primary and re-flash the corrupted software. Prior to this, AMD did not support writable BIOSes, but this implementation was an interesting change.

The rest of the ports are located on the I/O panel. The reference video card is equipped with two dual DL-DVI connectors, two Mini-DisplayPort ports and HDMI. It is worth noting that the 6970 model supports a maximum resolution of up to 2560 x 1600 on 3 monitors. And with a multi-stream hub using DisplayPort 1.2 mini-connectors, the card is capable of serving up to 6 displays. Like the 5870, the back of the device is covered with a metal plate. Although there are no components that require protection, this solution is well received by users, as it allows you to take a card without worrying about touching sharp contacts.

Although the overall dimensions of the HD 670 are almost identical to the 5870, according to users, AMD's bulky design is not entirely successful when it comes to working in CrossFire mode with 2 cards installed side by side. The 5870 shroud protruded slightly in the center, keeping the vents from being blocked by adjacent boards. The 6970 doesn't have that luxury, and it's possible to practically isolate the top card depending on how it's installed. As a result, the temperature rises, but not to critical values. AMD should learn from its competitors from NVIDIA and allocate more space around the fan, which would allow it to "breathe" freely. Users strongly recommend installing graphics cards with as much space as possible between them, if the motherboard and case allow it.

Performance

According to user feedback, the Futuremark 3DMark 11 test showed that when performing extreme tests, the AMD Radeon HD 6970 is 7% faster than the GTX 570 (1821 points vs. 1697). If this advantage persists in gaming tests, then this may play into the hands of the manufacturer. According to the same test, the HD 6970 is 7% slower than the GTX 580 (1962 points), 27% behind the HD 5970 (2506) and 16% faster than the HD 5870 (1572 points).

Dirt 2

This game has a fantastic built-in test that measures actual performance very accurately. Users ran Dirt 2 in DirectX 11 mode with 4xAA enabled and maximum image quality settings. The Radeon 6970 averaged 77 fps at 1920 x 1200, making it only 10% faster than the HD 5870. More importantly, the HD 6970 was 21% slower than its main competitor, the GTX 570. Other pricier models also lagged behind. The HD 5970 was 23% faster than the 6970, while the GTX 580 had a 33% advantage. Dirt 2 proved to be merciless for the HD 6970.

F1 2010

With the first update, the game got a great built-in benchmark that measures the card's performance very accurately. Users experienced F1 2010 in DirectX 11 mode with 8xMSAA enabled and better visual settings. It should be noted that they couldn't get CrossFire to work correctly with the latest drivers, so the HD 6970 couldn't take the lead. The F1 2010 results are not surprising as the HD 6970 is the second fastest graphics card tested at 1920 x 1200, only 9% slower than the powerful GTX 580. The HD 6970 is only 5% better than the 5870. The results in this game are a bit off the mark, though, as the HD 6970 leads the GTX 570 by a convincing 22%.

COD: Modern Warfare 2

To evaluate the performance of the video card in this game, users used the Fraps application. The maximum game quality settings were accompanied by 4xAA and 60-second gameplay recording. The Radeon 6970 tested at 73fps, which is 1fps faster than the GTX 480. This is close to the performance of the GTX 570 (74fps) and 16% behind the faster GTX 580. Compared with the dual processor Radeon 6970, the model was 19% slower, although it delivered 30% faster frame rates than the HD 5870.

COD: Black Ops

Fraps also helped benchmark this single player game. Users measured the frame rate during 1 minute of gameplay of the first single level (Operation 40) with maximum visual effects, including 4xAA. The Black Ops results showed that the Radeon 6970 is 5% behind the GTX 570 and 13% behind the GTX 580. predecessor by 24%.

Energy consumption

Compared to the HD 5870, the 6970 draws 16.5% more power, but considering it's also 24% faster, owners find it to be a better graphics card overall. Compared to the HD 5970, the 6970 is 3% more fuel efficient, but is 15% slower on average and 26% slower in Crysis Warehead, which was used in our stress test. The Radeon HD 6970 consumes about the same amount of power as the GTX 570, consuming 1% more under extreme loads and 4% less at idle. Given that both graphics cards deliver the same performance, this time around it seems they are equally matched.

Heating temperature

According to user feedback, the HD 6970 gets quite hot during FurMark stress testing, reaching 90°C, which is slightly higher than the HD 5870's 87°C. On the other hand, it only gets to 81°C, which makes it significantly cooler under maximum loads.

Overclocking performance

The Catalyst Control Panel limits the maximum core frequency to 950 MHz, which the Radeon HD 6970 can handle without issue. This is quite a worthy indicator. The memory frequency can go up to 1450 MHz, although users are faced with stability issues and are forced to fall back to 1440 MHz. This 8% overclock results in a 7.6% performance increase when tested in COD: Black Ops, 7.8% in Crysis Warhead, and 7% in Battlefield Bad Company 2.

Conclusion

It seems that the HD 6900 and GTX 500 repeat each other. With a few key differences, of course. The NVIDIA product hit the market just in time and proved to be very effective against competing AMD models. The GeForce GTX 480 was the fastest single-processor graphics card, and the GTX 580 retained the crown, as it was beyond the power of the Radeon 6970. The price of the model corresponded to the cost of the GTX 570, and the HD 6950 occupied the 580th niche. lead the way in performance by releasing 2 cards in the series, codenamed Antilles.

The Radeon HD 6970 goes hand in hand with the GTX 570 and has been set at $369 MSRP, while the GTX 570 debuted at $349. Although they are two very different models, they perform similarly in averages across many games tested by users at 1920 x 1200 resolution. For the most part, the difference is very small. The Radeon 6970 is 20% faster than the GTX 570 in some cases, and about 20% slower in others. In terms of power consumption, both cards are also very close, demonstrating similar levels of operational efficiency. Therefore, the HD 6970 is a good alternative to the GTX 570. It's not an easy choice between the two cards, but either one is the right one given their price and performance.

Compared to other models such as the GTX 580, the 6970 is 15% slower on average but costs 30% less, just like the GTX 570 which offers an even better price. The card that the HD 6970 effectively replaces is the HD 5870. Based on test data at 1920 x 1200, the HD 6970 is 24% faster on average and only 10% more expensive. In addition, the model's energy efficiency has also been improved. An increase in power consumption of 16% gives an additional 24% performance. It is worth noting that the Radeon 6970 looks slightly better than the GTX 570 in STALKER: Call of Pripyat and Aliens vs. Predator that use tessellation. Prior to its introduction, NVIDIA had a significant advantage here.

Overall, both graphics cards deliver exceptional levels of performance and value, making choosing between them both easy and difficult. But any decision will be correct.